Imagine having an AI teammate that not only writes code but tests, debugs, and iterates—all on its own. That’s exactly what OpenAI is delivering with its brand-new Codex AI agent, now live in research preview. 🎉
Built on the powerful Codex-1 variant of the o3 model, this coding agent is optimized for real-world software engineering and is available to ChatGPT Pro, Team, and Enterprise users. More access is coming soon for Plus and Edu subscribers too!
⚙️ What Is Codex AI and What Can It Do?
The new Codex AI agent isn’t just another autocomplete tool. It’s designed to independently complete software tasks, including:
- 🔧 Bug fixing
- 🧪 Writing and running tests
- 🚀 Adding new features
- 📁 Working directly with GitHub repos
All within 1 to 30 minutes, right in your ChatGPT sidebar!
🔍 How It Works:
- Runs in a secure sandboxed VM hosted in the cloud
- Can test its own code and retry until it works
- Operates in the background, so your computer stays free
- Manages multiple tasks at once
Think of it as a virtual developer assistant that keeps working while you focus on other things.
🧠 Smart Interface & Workflow
Assigning a task is simple:
- Type your prompt in ChatGPT.
- Choose “Code” (for implementation) or “Ask” (for insight).
- Track progress in a sidebar task list, just like a project dashboard.
OpenAI’s Josh Tobin says the vision is to make Codex a “virtual teammate” that can finish tasks that would normally take hours—or even days.
🔐 Safety First: Secure, Air-Gapped Design
Codex prioritizes security with an air-gapped virtual machine, meaning:
- ❌ No internet access
- ❌ No third-party APIs
- ✅ Reduced risk of misuse
- ✅ Refuses malicious requests by design
It inherits the same safety mechanisms from the o3 model and avoids generating code that could cause harm.
⚠️ Limitations to Keep in Mind
As smart as Codex is, it’s not perfect.
- 🐞 Still prone to debugging errors
- 📉 Microsoft found even top-tier AI models struggle to fix broken code reliably
- 🔄 May need human review for complex logic
It’s powerful—but not a complete developer replacement (yet!).
💡 Codex CLI Gets a Power Boost Too
OpenAI is also rolling out an update to Codex CLI, its terminal-based open-source coding agent. The CLI now uses the o4-mini model, finely tuned for engineering tasks.
📊 API Pricing:
- $1.50 per million input tokens
- $6.00 per million output tokens
Available via OpenAI’s API starting today!
⏳ Coming Soon: Rate Limits and Credit Options
While users currently enjoy generous access, OpenAI will soon introduce rate limits, with paid credits available to expand usage.
🚀 Final Thoughts: A Glimpse Into the Future of Software Engineering
Codex marks a huge leap toward autonomous AI development assistants. From startups to solo developers, this agent could significantly speed up workflows and reduce repetitive tasks.
🧠 Whether you need a test suite written or a legacy bug squashed, Codex is shaping up to be your AI-powered teammate.